IJSTE - International Journal of Science Technology & Engineering | Volume 2 | Issue 10 | April 2016

ISSN (online): 2349-784X

Segmentation of Histopathological Images using

Fast Fuzzy C-Means Approach
S.Kasthuri
UG Scholar
Department of Electronics and communication Engineering
Christian College of Eng. & Tech Oddanchatram
TamilNadu 624619 India

J. Lilly Pushpam
UG Scholar
Department of Electronics and communication Engineering
Christian College of Eng. & Tech Oddanchatram
TamilNadu 624619 India

K.Mahalakshmi
UG Scholar
Department of Electronics and communication Engineering
Christian College of Eng. & Tech Oddanchatram
TamilNadu 624619 India

Viviyenmol M D
UG Scholar
Department of Electronics and communication Engineering
Christian College of Eng. & Tech Oddanchatram
TamilNadu 624619 India

P. Tharics
Assistant Professor
Department of Electronics and communication Engineering
Christian College of Eng. & Tech Oddanchatram TamilNadu 624619 India

Abstract
Segmentation accuracy determines the success and failure of histopathological application. The count of the mitotic cells is an
important part of the cancer grading system. Segmentation of mitotic cells from the histopathological images is a crucial task for
the detection of cancer cells. Pathologist performed this task by a tedious job and result was not accurate. Many techniques have
been implemented for the automatic detection of cells including morphological, neural network etc. There are many challenges in
the detection and classification is based on accuracy of cell segmentation. We propose a fast robust method for cell segmentation
in histopathological images. We employee fast fuzzy c-mean algorithm. This method is stable and faster and we get more accurate
segmentation and easy detection a cancer cell.
Keywords: Histopathological images, segmentation, fast FCM, cancer grading system.
________________________________________________________________________________________________________
I.

INTRODUCTION

Image Processing is a method to convert an image into digital form. It performs some operation. In order to get an improved image
or to take out some useful information on it. It is a type of signal dispensation in which input is an image, like video border or by
photograph and output may be image or characteristics associated with that image. Usually image process system includes the
images as two dimensional signals. While applying already set signal processing methods are to them. The two types of image
processing methods are analog and digital image processing.
Medical imaging is the method and process of being created visual representations of the inside the body for clinical analysis
and medical intrusion, as well as visual representation of the some organs or tissues physiology.
Histopathological images are most widely used in breast cancer detection. Histopathology is being referred to the microscopic
examination of tissues. In order to studied the symptom of diseases. Fig1. Shows the some examples of histopathological images.
Image segmentations are in the process of converting an image into multiple parts. This is used to identify the objects or other
related information in digital images. There are many techniques used to perform image segmentation including the thresholding
method such as Otsus method. Image segmentation is the process of assigning a label to every pixel in an image such that same
labels with pixels share particular characteristics. Types of image segmentation methods are based on watershed transformation,
model based segmentation, multi-semi-automatic segmentation scale segmentation, trainable segmentation, segmentation
benchmarking.

Fig. 1: Histopathological images

All rights reserved by www.ijste.org

282

Segmentation of Histopathological Images using Fast Fuzzy C-Means Approach

(IJSTE/ Volume 2 / Issue 10 / 055)

II. RELATED WORKS

Angshuman Paul and Dipti Prasad Mukherjee, Senior Member, IEEE (2015) introduced fast and accurate approaches for automatic
mitotic detection for histopathological images using morphological scale space [1].
Pin Wang, Xianlling Hu, Yongming Li, Qianqian Liu and Xinjian Zhu (2015), proposed the automatic quantitative image
analysis method for BCH images are proposed. For the nuclei segmentation, top-bottom hat transforms are applied to improve the
enhanced image quality. Wavelet decomposition and multi-scale region-growing (WDMR) is combining to obtain regions of
interest (ROIs) thereby realizing the precise location. A double-strategy is splitting model (DSSM) containing adaptive
mathematical morphology and curvature scale space (CSS) for corner detection methods are applied to splitting overlapped cells
used for better accuracy and robustness [12].
AnujKumar Singh and Bhupendra Gupta (2015) introduce a simple and easiest way for detection of cancerous tissues in a
mammogram. Detection phases are followed by segmentation of the tumor region in a mammogram image. Our approach uses
simple image processing methods such as averaging and thresholding. We introduce a Max-Mean and Least-Variance technique
for tumor detection [2].
Danilo cesar pereira, Rodrigo pereria ramos, Marcelo Zanchetta do nascimento (2014) proposed an artifact removal algorithms
are first implemented by following image denoising and gray-level enhancement method based on wavelet transform and wiener
filter. Finally, a method using for detection and segmentation of masses using multiple thresholding, wavelet transform and the
genetic algorithms are employed in mammograms which are randomly selected from the Digital Database for Screening
Mammography (DDSM) [5].
Humayun Irshad, Sepehr Jalali, Ludovic Roux, Daniel Racoceanu, Lim Joo Hwee, Gilles Le Naour, Frederique Capron(2014)
proposed an approach that assists pathologists in automated mitosis detection and counting. The proposed methods, which are
based on the most favorable texture features combination, examine the separability between different channels of color space [6].
III. PROPOSED METHOD
During four phases of mitosis, the shape of mitotic cells appears with different. The detection and classification must be based on
some features that remain unchanged. The intensity pattern of mitotic cells at different phase remains invariant. But this factor
intensity can be a good feature for classification of mitotic cell only if the cells are accurately segmented. In proposed method
(Fig: 1), we are doing up to segmentation part.

All rights reserved by www.ijste.org

283

Segmentation of Histopathological Images using Fast Fuzzy C-Means Approach

(IJSTE/ Volume 2 / Issue 10 / 055)

Fig. 2: Pipeline of the Proposed Method

Pre-Processing:
Preprocessing images commonly used to reduce the low-frequency background noise, the intensity of normalize the individual
particles images, removing reflections, and masking portions of images. Preprocessing is the technique of being enhanced the data
images in to computational processing.
Wiener Filter:
Non causal Solution
G(s) =

Sx,s(S) s
e
Sx(S)

(1)

Where S is spectra provided that g (t) is optimal, then the minimum-mean square equation reduces to

E ( 2 ) = (0) - (), (+) d

(2)
The solution of () is the inverse two-sided laplace transform of. ().
Casual solution
()
() = +
(3)
()

where,

() is the causal part of

,()
()

(that is, the part of this fraction. It consist of a positive time solution below that the inverse

laplace transform )
+ () is the causal solution of () (i.e., the inverse laplace transform of + () is non-zero only for t 0
()is the anti-causal component of () (i.e., the inverse laplace transform of ()is non-zero only for t < 0
This general formula is difficult and more detailed explanation. To write down the solution in the specific case, to follow the these
steps
1) Start () is the spectrum of () onal form and factor. It is the causal and anti-causal components:
() = + () ()
(4)
It contains all zeros and poles in the left half plane (LHP) and the zeroes and poles in the right half plane (RHP). This is called
as Wiener-Hop factorization
2) Divide ,() by () and write out, the result as a partial fraction expansion.
3) Select only those expressions in this expansion having poles in the LHP. Called these terms are ()
4) Dividing the () by+ (). The results are the desired filter transfer function of ().
In order to receive the coefficients of the wiener filter, consider the signals w[n] being fed to a wiener filter of order N and with
coefficients. The {0 }output of the filter is denoted as x[n] which is given by the equation as
[] =
(5)
=0 [ ]
The residual errors are denoted by e[n] and is defined as e[n] = x[n] s[n]. The Wiener filters are designed to minimize the
mean square error MMSE which can be stated concisely as follows:
= arg min [ 2 []]
(6)
Where [. ] denotes the expectation operator. In this case, the coefficients may be complex and may be derived for the case
where w[n] and s[n] is the complex. For, the following considers are only the case where all these quantities are real.
Image Enhancement:
The image enhancements are used to improve the interpretability or perception of information in the images for human viewers,
or to provide `better' inputs for other automated image processing methods. Image enhancement methods can be used to divide the
two broad categories. Spatial domain methods, which operate directly in the pixels, and methods of frequency domain, which
operates in the fourier transform of an image. Unfortunately, there being no general theory for determine what `good image
enhancement is when it comes to human perception. If it is looking good, it is good! However, when image enhancement. The aim
of these image enhancements is to improve the interpretability or for human viewers information perception in an image, or to
provide `better' input for other automated image processing techniques. Image enhancement techniques can be used to divide the
two categories:
1) Spatial domain method, which operate directly on pixels.
2) Frequency domain method, which operates the fourier transform of an image.
Unfortunately, there being no general theory for determine what `good image enhancement is when it comes to human
perception. If it looks good, it is good! However, when image enhancement techniques pre-processing tools are used for other
image processing methods, and quantitative measures can be determine which techniques are most appropriate.
Smoothing:
In image enhancement, we are applying smoothing in order to reduce noise and to reduce preparing images for further processing
such as segmentation. Smoothing otherwise called as blurring. In smoothing we are using a number of filters. Here we are using
the median filter to reduce the noise in histopathological images.

All rights reserved by www.ijste.org

284

Segmentation of Histopathological Images using Fast Fuzzy C-Means Approach

(IJSTE/ Volume 2 / Issue 10 / 055)

Smooth the image with a Gaussian filters to reduce noise and unwanted details and textures.
(, ) = (, ) (, )
(7)
where
=

1
2 2

exp (

2 +2
2 2

(8)

Compute gradient (, ) of using any of the gradient operators (Roberts, Prewitt, etc) to get:
2 (, ) + 2 (, )
(, ) =

(9)

and,
(, ) = tan1 [ (, )/ (, )]
(10)
(, ) (, ) >
(, ) = {
0

(11)
Segmentation Of Histopathological Images:
Image segmentations are the process of converting a digital image into multiple segments. The goal of segmentation is used to
simplify or change the representation of an image into something that is most meaningful and easily to analyze.
Fast Fuzzy C-means (FFCM) algorithm
FFCM assigns c membership grades to every pixel. However, updating membership matrix with c n members are the time
consuming procedure. In FFCM, centroids are updated by fuzzy a membership which needs more time. To reduce the time and to
reduce the amount of computations in FFCM, a hard membership can be assigned to pixels for updating cluster centers in each
iteration step. However, segmentation will be a fuzzy procedure. By applying the hard membership of new algorithms to be update
centroids are proposed as following steps:
1) For p-dimensional input data, rearrange u ij to d1d2 matrix; where d1, d2 are input dimensions.
2) Set new fuzzy membership as u ij * and label matrix as
L = {L1, L2, Lc}; where Lk is label matrix of the k-th cluster in current iteration.
3) Set all data point which is corresponding to Lk label matrix as Ik.
4) Define Ik=I1, I2 ...Ik nck for the k-th cluster, where nck is the number of data points in kth cluster.

=1

(12)

th

Centroid of k cluster by equation

The new fuzzy membership and cost function can be calculated the following equations:
1

=
(13)
2/(1)

)

=1(

( )

= (
)

=1 =1

(14)
The FFCM algorithm can be summarized as following:
1) Select the number of clusters(c) and fuzziness value (m=2), initialize by small values.
2) Update the new membership matrix U* by Eq.
3) Update cluster center matrix V* by Equ.
(+1)
()
4) Repeat steps 23 until
where is a small positive constant.
Fast FCM algorithm is a fast and robust method for image segmentation. It is easily computed, better segmentation and better
robustness. This is more accurate and faster than other segmentation methods.
IV. SIMULATION RESULTS AND DISCUSSIONS
For this process, we are used to MATLAB 9a for simulation.

All rights reserved by www.ijste.org

285

Segmentation of Histopathological Images using Fast Fuzzy C-Means Approach

(IJSTE/ Volume 2 / Issue 10 / 055)

We perform segmentation using many methods including watershed algorithm, Otsus method, Fuzzy K-means method, texture
method, FCM, Fast FCM, etc. Among all these techniques fast FCM gives more accurate segmentation in histopathological images
(Fig 3). From over all discussion and analysis fast FCM provide high segmentation accuracy.

S.NO

Input histopathological images

Segmentation output using fast

FCM

Outlined input histopathological images

4
Fig. 3: Segmentation of histopathological images using fast FCM

So, from fast FCM segmentation output we can further classify the mitotic and non-mitotic cells and this leads to improved
cancer detection.
V. CONCLUSION AND FUTURE WORK
In this paper, we propose a fast and robust method for cell segmentation of histopathological images. We use fast fuzzy c-means
algorithm for the accurate segmentation of breast cancer cells. We apply a wiener filter in the preprocessing stage for the removal
of noise of the histopathological images and finally accurate segmentation will be obtained from fast FCM overall study and
analysis shows significant improvement over all other existing method for segmentation of histopathological images. In future,
We are extending this proposal for processing cancer cells in various parts of the body using soft computing techniques
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]

Angshuman Paul and Dipti Prasad Mukherjee, Mitosis detection for Invasive Breast Cancer Grading in Histopathological Images, IEEE Transs. on image
processing. VOL. 24, NO.11, NOVEMBER 2015.
AnujKumar Singh and Bhupendra Guptha A Novel approaches for breast cancer cells detection and segmentation in a mammogram cells, Vol. 54 Aug
2015
Barghout, Lauren, and Lawrence W. Lee. Perceptual information processing system." Paravue Inc. U.S. Patent Application 10/618,543, filed July 11, 2003.
C. E. Shannon, A mathematical theory of communication, ACM SIGMOBILE Mobile Computer. Rev Vol. 5, no. 1, pp. 3-55, 2001.
Danilo Cesar Pereira, Rodrigo Pereria Ramos, Marcelo Zanchetta do Nascimento, Segmentation and Detection of Breast Cancer cells in mammograms
combining wavelet based analysis and genetic algorithm, April 2014, Vol. 114, issue 1.
Humayun Irshad, Sepehr Jalali, Ludovic Roux, Daniel Racoceanu, Lim Joo Hwee, Gilles Le naour, frederique capron,Automated Mitosis Detection using
texture, SIFT features and HMAX biologically inspired approaches, Vol. 4, March 2013, pp. 12.
Kashanipour, N Milani, A. Kashanipour, H. Eghrary," Robust Color Classification Using Fuzzy Rule-Based Particle Swarm Optimization", IEEE congress
on image and signal processing, Vol. 2, pp. 110-114, May. 2008.
L. Breiman, Random forests, Mach. Learn, Vol. 45, no. 1, pp. 5-32, 2001.
L. Xu, C. Lu, Y. Xu and J.Jia, Image Smoothing Via L0 Gradient Minimization, ACM Trans. Graph, Vol. 30, no. 6, 2014, Art ID 174.
M. held ET al., Cellcognition: Time-resolved phenotype annotation in high-throughput live cell image, Nature Methods, Vol. 7, no. 9, pp. 747-754, 2010.
M. Seyedhosseini and T. Tasdizen, Multi-class Multi-scale series contextual model for image segmentation, IEEE Trans. Img. Process, Vol. 22, no. 11,
pp. 4486-4496, Nov. 2013.
Pin Wang, Xianlling Hu, Yongming Li, Qianqian Liu and Xinjian Zhu,Automatic cell nuclei segmentation and classification of breast cancer histopathology
images, Vol.122, May 2015.
Roobottom CA, Mitchell G, Morgan-Hughes G; Mitchell; Morgan-Hughes. "Radiation Reduction strategies in cardiac computed tomographic angiography".
Clin Radiol 65 (11): 85967, Nov. 2010

All rights reserved by www.ijste.org

286